Sarcasm detection in political speeches using recurrent neural networks

Abstract

Sarcasm detection is a challenging task in natural language processing (NLP) that has received significant attention in recent years. Sarcasm is a form of indirect speech in which the speaker says the opposite of what they mean. It can be used to express a variety of emotions, such as humour, irony, or contempt. Sarcasm is often difficult to detect, especially in written text, because it often relies on context and the speaker's intent. Recurrent neural networks (RNNs) have been shown to be effective in sarcasm detection, but there is still room for improvement. In this work, we propose a stacking and weighted average ensemble model using simpleRNN, LSTM, and GRU as base models for sarcasm detection. The news headline dataset was used in the study. The dataset contains sarcastic and non-sarcastic labels for the headlines, and contains a total of 55325 headlines, the dataset is split into 80% (44260) testing and 20% (11065) validation. The aim of this study was to develop a model to detect sarcasm in political speech using Recurrent Neural Networks, incorporating sarcasm detection into sentiment analysis for political text can significantly enhance the accuracy and depth of sentiment understanding. The results suggest that the ensemble models outperform individual neural network models, with the two-level stacking model achieving the best overall performance.